How I used Mastra to build a prize-winning RAG agent

In 2025, the surge in AI agents has captured the tech world’s attention, leading to the development of innovative frameworks like Mastra, an open-source TypeScript tool for building AI agents. The author delves into Mastra through participation in a hackathon, creating an AI agent that answers questions about video transcripts using natural language, leveraging the Retrieval-Augmented Generation (RAG) technique. Mastra, founded by Sam Bhagwat, Abhi Aiyer, and Shane, enhances AI development with components like agents, tools, workflows, memory, and RAG, all built on Vercel's AI SDK to enable dynamic model routing. The AI agent integrates semantic search and full-text search for retrieving transcript information, enriched with metadata for accuracy and context, and uses OpenAI’s text embedding model for semantic understanding. The project highlights the importance of prompt engineering and explores the practical application of AI tools in real-world tasks, offering insights into the evolving landscape of AI agent development.